Table of Contents

Monetization for mobile game

Data upload and preparing it for analysis

The logs contain information about 13,576 players. Only 1 duplicate, can be neglected.

The number of events with the type of implemented project "development of orbital assembly of satellites" coincides with the number of events containing information about the completion of the project. According to the team's comments, the omissions in the "project_type" column are due to the mechanics of data collection.

In total, 4 advertising channels were used, the date range corresponds to the declared cohort data with a time lag between the user's involvement and the start of using the application in 1 day.

All user IDs are unique.

We have prepared the data for analysis by bringing it to the desired format and checking for omissions/duplicates.

Exploratory data analysis

Average number of events per user

The number of unique user IDs in the session data coincides with the number of IDs in the data on the attraction channels.

Number of observations depending on time

The graph shows how the number of users increased while the advertising campaign was going on until May 10, after that there is a sharp jump down in the number of visits (apparently, some users leave after the first use), then users already flow away gradually.

The distribution of the number of observations about the completion of the first level is close to normal. The majority of users completed the first level between May 13 and May 21 - it takes users several days to complete the level.

The number of users in the game depending on the day of the week

On every day of the week, users are approximately equally active.

Choosing the type of task by players

Distribution of observations by event type

The most frequent type of event in the data is the construction of a building.

Percentage of players who completed the first level

At the moment, the creators of the game plan to show ads only on the screen with a choice of buildings: but most of the users complete the game by defeating another player. It is possible that some of the users who were dragged out by the game and moved to the second level will not see ads at all.

Let's see if we have players who were able to move to the second level without having a single building in the logs.

There are no players in the cohort who complete the level without trying to build a single building, therefore, showing ads on the screen with the choice of the building object is a working idea. Let's see which buildings are chosen for construction by the players who have completed the level.

Apparently, in order to complete the project, the player needs to build all 3 types of buildings.

Adding calculated data about the time of passing the game

We will find for each user the date of the first interaction with the game, as well as the date of the last event.

On average, the difference between events in the cohort is 10 days, and is only a few hours different from the median. At the same time, there are players who have used the game for 30 days, and there are also those for whom there is only 1 event. Let's look at the example of 1 user, is there a difference between how the time of the "object is built" event is set and the event about the fact of passing the first level.

The user selected all three types of objects for construction several times, while about 10 hours passed between the events "Project completed" and "First level passed". For further analysis, the moment of passing the first level will be considered the event "finished_stage_1".

The time spent in the game by users who have passed the first level

The average time to complete the level is quite variable, the spread is approximately from 9.5 hours to 742. The average passage time is 285 hours, i.e. almost 12 days. Let's look at the distributions for the two ways of passing the level.

Analysis of the passage time of the game depending on the method of passage

Both distributions are close to normal, which will allow us to use the T-criterion to compare the averages further. Let's look at the descriptive statistics for two samples.

In general, there are many more people in the cohort who completed the level with a victory over another player, and on average it took them less time (both on average and median), but the variability of the time of passing through the victory over the player is much higher.

The time spent in the game by users who have not passed the first level

There are users who fall off immediately (the minimum time spent in the game is 0 hours).

Number of events for a user who has not passed the first level

Users fall off at about step 9.

Events by building types for players who have not moved to the 2nd level

Building a research center is the most unpopular type of task. Let's look at the ratio of each type of building to the buildings of all users in the entire cohort.

It is difficult for players to build a research center: at this stage, most of the players fall off.

Marketing

Distribution of users by channels of attraction

Most of the users came through Yandex Direct, least of all through advertising on YouTube. The two most popular channels brought about 60% of users.

Advertising costs by channel

The daily budget on the platform was about 300 USD on the three most popular platforms, while on YouTube it was about 150 USD.

Let's look at the average cost of attracting through channels.

Advertising costs decreased gradually throughout the week.

The cost of attracting a user

In general, by cohort
By channels

The most expensive channel by the average cost of attraction is facebook_ads, the least expensive is youtube_channel_reklama. The cost of attracting users via instagram_new_adverts and facebook_ads channels is higher than the average for the cohort, for yandex_direct and youtube_channel_reklama channels is lower.

By days

It can be seen that marketing tried to keep the cost of attracting a client approximately the same throughout the entire advertising campaign.

EDA summary

User behavior

  1. 42.85% of the players in the cohort completed the first level, while 29.10% did it through defeating another player.

  2. Building a research center is the most unpopular type of task, at this stage the most players fall off.

  3. The average number of events per user who has not passed the first level is 8,7. The passage of the game through the completion of the project takes longer than through the victory over another player. We will check this in the future using the statistical method.

  4. All players in the cohort chose an object to build - even those who completed the level through defeating another player. Therefore, with the current monetization model, there is not a single player who could not see the advertisement.

Marketing

  1. We analyzed the distribution of channels by the number of attracted customers. Most of the customers came through yandex_direct (35% of users), 2 times more than through youtube_channel_reklama.

  2. We calculated the average CAC in the whole cohort, in the section of channels and days of the advertising company. The cost of attracting a user was about the same. The facebook_ads channel turned out to be the most expensive, the most effective by this indicator - youtube_channel_reklama.

Monetization model

Let's evaluate the existing monetization model. It is supposed to show ads on the screen with a choice of buildings, for 1 such display we earn 0.07 CU. Let's calculate how much we would have earned in the cohort under study in a month by the number of events - completed buildings:

Considering that the cohort was not shown ads and not all users left as quickly as they could, our margin was 17.8%. However, some users could see the screen with the choice of building, but not complete the construction of the building, and the event did not get into the logs.

The average number of events per user in the logs is about 10 (9.99 pcs.). On average, attracting a user costs 0.56 yandex units, for 1 impression we earn 0.07 yandex units, therefore, the user must see the ad at least 8 times to pay off.

Let's see what margin is possible in the current cohort if does not show ads for the player's first building, but instead shows it on all screens of the level. Such a model can work, because after the first construction, the player will feel the taste of the game and will want to continue building other objects.

The margin turns out to be less than with a conservative assessment of the current monetization model. This option can be a working one if it is possible to significantly reduce the CAC by optimizing the marketing strategy. Let's assume that we will still show ads on the screen with the completion of the construction of the 1st object:

We can use this model.

Testing statistical hypotheses

The hypothesis about the difference in the passage time of the level between users who finish the level through the implementation of the project, and users who finish the level with a victory over another player

This is a hypothesis about the equality of two averages, let's use the t-test.

H0 - the average passage time between users who complete a level through the implementation of the project and users who complete a level by defeating another player does not differ.

H1 - the average passage time is different.

There is a difference between the two aggregates.

Hypothesis about the difference in the time of passing the game between users who installed the game on a day off and those who came during the week

Perhaps before the weekend and on weekends, people have less stress and distractions, and they are ready to spend more time in the game, which means they will pass the first level in fewer days.

This is a hypothesis about the equality of two averages, let's use the t-test.

H0 - the average transit time between users who logged in to the app for the first time on a weekend and users who logged in for the first time on weekdays does not differ.

H1 - the average passage time is different.

There is a difference between the two aggregates.

The difference is almost a day.

Conclusion and recommendations on the monetization model